In recent years, it has become possible to send/receive electronic mail not only on a personal computer but also on a portable terminal. Furthermore, not only documents, but also a sender's portrait have been widely attached to the electronic mail. When creating such a portrait by using a conventional method, it is necessary for the user to manually make selections from prepared parts of several tens or more kinds, such as face outline parts, face parts and hairstyles, and combine them all together. In this case, a large number of combinations are available, and, as a result, a great workload is imposed on the user in an attempt to create a desired person's portrait. For this reason, there has been a strong demand for a device which can easily and automatically generate a person's portrait using an image of that person acquired by a camera. In order to create a person's portrait, such a device needs to automatically carry out a sequence of processes of extracting the face portion from the image, detecting eyes, mouth, etc., and replacing the desired parts with already prepared desired templates.
As one example of the related technique, Japanese Laid-Open Patent Application No. 255017/1998 (Tokukaihei 10-255017) discloses a method and device for generating a person's portrait. In this method, with respect to the face area detection, (R, G, B) values of skin color are preliminarily set to a specified range, and pixels having the corresponding color are marked, and clustered so as to carry out the detection, while the face parts detection is carried out by template matching processes. However, with respect to the face area detection,                (1) the face color is influenced by illumination conditions, resulting in a difficulty in detecting the face area in some cases;        (2) the face color is not located within the above-mentioned range of the (R, G, B) values due to differences in human races, etc.; and        (3) when pixels having a color similar to the face color exists on the background, it is sometimes difficult to detect the face area.        
Moreover, with respect to the face parts position detection, conventional problems are that,                (4) when the number of face-part templates is increased so as to match various face parts contained in an image, the amount of calculations increases due to increased collation processes, resulting in degradation in the execution speed; and        (5) a dark shadow tends to appear on the periphery of a face part in an image depending on illumination conditions, resulting in a difficulty in the template matching.        